Summary Our proposal describes network based approaches for the analysis of data from metabolomics studies. The speci?c aims of this proposal include: Aim 1: Variable selection methods in metabolomics studies, incorporating metabolite dependence and external pathway information. We propose a Bayesian variable selection approach to incorporate both a partially observed, external pathway network and a data-driven partial correlation network. Aim 2: Models to identify differential metabolic networks that characterize groups within a study, and addi- tionally detect subcomponents with group-speci?c associations with an outcome. When metabolic networks differ according to groups, exposure levels (e.g. treatment) or other factors, our proposed framework will provide an approach to identify group-speci?c networks as well as subcomponents that are associated with outcome, in a possibly group-speci?c manner. Aim 3: Methods to identify metabolite subnetworks that collectively mediate the relationship between an ex- posure and an outcome. We propose a two-phase analysis framework involving (1) Detection of metabolite subnet- works enriched for association with the outcome; and (2) Estimation of the magnitude of the indirect effects mediated by metabolite subnetworks. Application to testing clinical hypotheses in the WHI, NHS and HAPO metabolomics studies: Using methods developed in Aims 1, we will identify metabolites and modules associated with risk of stroke in the NHS and maternal metabolomic markers of newborn adiposity in the HAPO study. Using methods in Aim 2, in the WHI, we will identify metabolic subnetworks that change due to initiation of hormone therapy (estrogen, progestin plus estrogen, placebo) within age groups, with treatment/age dependent modules associated with subsequent risk of CHD; in the HAPO study, detect maternal metabolite networks that differ between mothers of boys versus mothers of girls and sex-speci?c subcomponents that inform sex-related differences in newborn body composition related to maternal glycemia during pregnancy. Aim 3 methods will be applied to detect metabolite subnetworks that potentially mediate the association of exposures such as dietary score and risk of CHD in the WHI; and maternal glucose during pregnancy and newborn adiposity in HAPO. IMPACT: Signi?cant federal investment has been made into research of the metabolomic underpinnings of complex disorders, such as through the NIH's Common Fund Metabolomics program. Our interdisciplinary team proposes to develop and apply new statistical models to effectively mine rapidly growing metabolomics data sources to elucidate the etiology of complex disorders such as CHD, stroke and maternal glycemia during pregnancy as it relates to newborn size at birth.